Fig. 1. Experimental setup of φ-OTDR. (a) Knocking with a hammer; (b) wind blowing. AOM, acoustic-optic modulator; EDFA, erbium-doped fiber amplifier; LO, local oscillator; BPD, balanced photodetector; DAQ, data acquisition card.
Fig. 2. (a) Developed prototype φ-OTDR instrument; (b) deployed fiber in the experiment.
Fig. 3. Flow chart of DSP.
Fig. 4. Time-domain waveform of vibration signal. (a) Knock around the fiber with a hammer; (b) wind blowing; (c) background noise.
Fig. 5. Temporal-spatial image and time-frequency image for vibration events. (a), (d) Knock around the fiber with a hammer; (b), (e) wind blowing; (c), (f) background noise.
Fig. 6. Network structure of ResNet50.
Fig. 7. Residual blocks of ResNet50.
Fig. 8. (a) Classification accuracy curve and (b) loss curve of training.
Fig. 9. Confusion matrix of (a) temporal-spatial image and (b) time-frequency image.
Event Type | Knocking | Blowing | Noise |
---|
Temporal-spatial training set | 599 | 804 | 888 | Temporal-spatial validation set | 421 | 396 | 312 | Total | 1020 | 1200 | 1200 | Time-frequency training set | 845 | 804 | 641 | Time-frequency validation set | 355 | 396 | 379 | Total | 1200 | 1200 | 1020 |
|
Table 1. Composition of the Data Set
| Precision | Accuracy | Average Training Time/s |
---|
Knocking | Noise | Blowing |
---|
Temporal-spatial image | 99.64% | 98.98% | 99.98% | 99.49% | 219 | Time-frequency image | 95.51% | 99.18% | 98.19% | 98.23% | 216 |
|
Table 2. Comparison of Training Results between Temporal-Spatial and Time-Frequency Images
Event Type | Recall | f1-Score |
---|
Temporal-spatial | Knocking | 99.76% | 99.76% | Noise | 99.36% | 99.52% | Blowing | 99.75% | 99.62% | Time-frequency | Knocking | 99.15% | 97.50% | Noise | 96.04% | 97.59% | Blowing | 100% | 100% |
|
Table 3. Comparison of Recall and f1-Score